|
Self-Driving Car Engineer
|
| Self-Driving Car Engineer Nanodegree, Udacity vamshi.krishna.prime@gmail.com |
|
Project: Finding Lane Lines on the Road¶
Table of contents:¶
- Introduction
- 1.1. Orientation:
- 1.2. Desired outcome:
- 1.3. Software requirements:
- 1.4. Key modules
-
2.1. Detect Lane Lines on images
- 2.1.1. Display the images in the test directory
- 2.1.2. Convert the images to grayscale (single channel))
- 2.1.3. Blur the grayscale image
- 2.1.4. Extract the edges in the image
- 2.1.5. Identify the region of interest
- 2.1.6. Mask the excess region
- 2.1.7. Draw Hough lines from the edges in the masked region
- 2.1.8. Superimpose the lines on the image to identify lane markings
Extrapolated Lane Line Detection
3.1. Extrapolated Lane Line Detection on images
- 3.1.1. Display the images in the test images challenge directory
- 3.1.2. Convert the images to grayscale (single channel))
- 3.1.3. Blur the grayscale image
- 3.1.4. Extract the edges in the image
- 3.1.5. Identify the region of interest
- 3.1.6. Mask the excess region in the images
- 3.1.7. Draw Hough lines from the edges in the masked region
- 3.1.8. Superimpose the lines on the image to identify lane markings
- 3.1.9. Filter the redundant lines in the images
- 3.1.10. Extrapoate lines with linear regression
- 3.1.11. Extrapoate lines on test images set
- 3.1.12. Linear Regression
1. Introduction¶
1.1 Orientation:¶
Leveraged various tools to identify lane lines on the road. Developed the pipeline on a series of individual images, and later applied the result to a video stream of different scenarios/environmental variables.
1.2 Desired outcome:¶
The initial goal of the project is displayed in the video clip "raw-lines-example.mp4" in the directory "examples"(also contained in this repository) to see what the output should look like after using the helper functions in the project. On successful identification of the land lanes, the result that looks roughly like the video clip "raw-lines-example.mp4" in the "examples" directory, improve the output by averaging and/or extrapolating the line segments detected to map out the full extent of the lane lines. You can see an example of the result you're going for, in the video clip "P1_example.mp4" in the "examples" directory. Ultimately the final output video clip displays just 2 lines; one line for the left side of the lane, and one for the right.
1.3 Software requirements:¶
This project uses Python 3 and is designed to be completed through the Jupyter Notebooks IDE. It is highly recommended that you use the Anaconda distribution to install Python, since the distribution includes all necessary Python libraries as well as Jupyter Notebooks. The following libraries are expected to be used in this project:
- NumPy
- pandas
- Matplotlib
- Seaborn
Additional Resources:
In addition to this jupyter notebook, there is a brief writeup in a separate file for reference, which can be either a markdown file or a pdf document.
Note:
If, at any point, you encounter frozen display windows or other confounding issues, you can always start again with a clean slate by going to the "Kernel" menu above and selecting "Restart & Clear Output".
1.4 Key modules in Lane Detection Pipeline:¶
Some of the Computer Vision techniques that are used to reach the desired outcome are listed below:
- Color selection
- Region of interest selection
- Grayscaling
- Gaussian smoothing
- Canny Edge Detection
- Hough Tranform line detection
Some OpenCV functions that are used in this project are:
cv2.inRange() for color selection
cv2.fillPoly() for regions selection
cv2.line() to draw lines on an image given endpoints
cv2.addWeighted() to coadd / overlay two images
cv2.cvtColor() to grayscale or change color
cv2.imwrite() to output images to file
cv2.bitwise_and() to apply a mask to an image
Check out the OpenCV documentation to learn about these and discover even more awesome functionality!
-------- End of `1. Introduction` --------
2. Detect Lane Lines¶
=================================
2.1 Detect Lane Lines on images¶
The goal is to piece together a pipeline to detect the line segments in the image, then average/extrapolate them and draw them onto the image for display (as below). Once we have a working pipeline, then we can move on to work on the video stream.
The orginal image used to detect land lines on the road |
→ |
The output detecting line segments using helper functions |
Import Packages¶
In [1]:
# import relevant packages
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np
import time
import math
import cv2
%matplotlib inline
Helper Functions¶
In [2]:
def grayscale(img):
"""Applies the Grayscale transform
This will return an image with only one color channel
but NOTE: to see the returned image as grayscale
(assuming your grayscaled image is called 'gray')
you should call plt.imshow(gray, cmap='gray')"""
return cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
# Or use BGR2GRAY if you read an image with cv2.imread()
# return cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
def canny(img, low_threshold, high_threshold):
"""Applies the Canny transform"""
return cv2.Canny(img, low_threshold, high_threshold)
def gaussian_blur(img, kernel_size):
"""Applies a Gaussian Noise kernel"""
return cv2.GaussianBlur(img, (kernel_size, kernel_size), 0)
def region_of_interest(img, vertices):
"""
Applies an image mask.
Only keeps the region of the image defined by the polygon
formed from `vertices`. The rest of the image is set to black.
`vertices` should be a numpy array of integer points.
"""
#defining a blank mask to start with
mask = np.zeros_like(img)
#defining a 3 channel or 1 channel color to fill the mask with depending on the input image
if len(img.shape) > 2:
channel_count = img.shape[2] # i.e. 3 or 4 depending on your image
ignore_mask_color = (255,) * channel_count
else:
ignore_mask_color = 255
#filling pixels inside the polygon defined by "vertices" with the fill color
cv2.fillPoly(mask, vertices, ignore_mask_color)
#returning the image only where mask pixels are nonzero
masked_image = cv2.bitwise_and(img, mask)
return masked_image
def draw_lines(img, lines, color=[255, 0, 0], thickness=2):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(img, (x1, y1), (x2, y2), color, thickness)
def hough_lines(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]), minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
draw_lines(line_img, lines)
return line_img
# Python 3 has support for cool math symbols.
def weighted_img(img, initial_img, α=0.8, β=1., γ=0.):
"""
`img` is the output of the hough_lines(), An image with lines drawn on it.
Should be a blank image (all black) with lines drawn on it.
`initial_img` should be the image before any processing.
The result image is computed as follows:
initial_img * α + img * β + γ
NOTE: initial_img and img must be the same shape!
"""
return cv2.addWeighted(initial_img, α, img, β, γ)
Read in an Image¶
In [3]:
#reading in an image
image = mpimg.imread('test_images/solidWhiteRight.jpg')
#printing out some stats and plotting
print('This image is:', type(image), ',', 'with dimensions:', image.shape)
plt.imshow(image);
# for single color channel image called 'gray', call as plt.imshow(gray, cmap='gray')
Test Images¶
Build the pipeline to work on the images in the directory "test_images"
A test to make sure the pipeline works well on these images before moving on to the videos.
In [4]:
import os
os.listdir("test_images/")
Out[4]:
Build a Lane Finding Pipeline on Images¶
Build the pipeline and run the solution on all test_images. Tune the various parameters, especially the low and high Canny thresholds as well as the Hough lines parameters. Make copies/output into the test_images_output directory, in order to use the images in the writeup report.
2.1.1. Display the images in the test directory¶
In [5]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.1. Display the images in the test directory\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.1. Display the images in the test directory.png', dpi=300, bbox_inches='tight')
2.1.2. Convert the images to grayscale (single channel)¶
As all the images in the test directory are colored (RGB channel), convert the images to single channel/grayscale for further computer vision operations.
In [6]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(gray, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.2. Convert the images to grayscale\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.2. Convert the images to grayscale.png', dpi=300, bbox_inches='tight')
2.1.3. Blur the grayscale image¶
In [7]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(blur_gray, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.3. Blur the grayscale image\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.3. Blur the grayscale image.png', dpi=300, bbox_inches='tight')
2.1.4. Extract the edges in the image¶
Extract the edges in the masked region by applying Canny Edge detection Computer Vision technique.
In [8]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.4. Extract the edges in the image\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.4. Extract the edges in the image.png', dpi=300, bbox_inches='tight')
2.1.5. Identify the region of interest¶
Identify the region of interest that is used for further operations. Take consideration of the hood of the car and eliminate it from the region of interest.
In [9]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# identify the region of interest
imshape = image.shape
x = [.55*imshape[1], .45*imshape[1], .15*imshape[1], .30*imshape[1],
.50*imshape[1], .70*imshape[1], .85*imshape[1], .55*imshape[1]]
y = [.60*imshape[0], .60*imshape[0], .90*imshape[0], .90*imshape[0],
.60*imshape[0], .90*imshape[0], .90*imshape[0], .60*imshape[0]]
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.plot(x, y, 'b--', lw=2)
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.5. Identify the region of interest\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.5. Identify the region of interest.png', dpi=300, bbox_inches='tight')
2.1.6. Mask the excess region¶
Keep the region of interest that is used for further operations and remove/darken the rest of the image to ease the identification of the lanes ahead of the vehicle.
In [10]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# identify the region of interest
imshape = image.shape
x = [.55*imshape[1], .45*imshape[1], .15*imshape[1], .30*imshape[1],
.50*imshape[1], .70*imshape[1], .85*imshape[1], .55*imshape[1]]
y = [.60*imshape[0], .60*imshape[0], .90*imshape[0], .90*imshape[0],
.60*imshape[0], .90*imshape[0], .90*imshape[0], .60*imshape[0]]
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges_masked, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.plot(x, y, 'b--', lw=2)
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.6. Mask the excess region\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.6. Mask the excess region.png', dpi=300, bbox_inches='tight')
2.1.7. Draw Hough lines from the edges in the masked region¶
Use Hough Line transformation to identify lines from the edges and draw the lines using various input parameters.
In [11]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 1 # distance resolution in pixels of the Hough grid
theta = np.pi/180 # angular resolution in radians of the Hough grid
threshold = 40 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 10 #minimum number of pixels making up a line
max_line_gap = 70 # maximum gap in pixels between connectable line segments
line_image = np.copy(image)*0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(edges_masked, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
# Create a "color" binary image to combine with line image
color_edges = np.dstack((edges, edges, edges))
# Draw the lines on the edge image
lines_edges = cv2.addWeighted(color_edges, 0.8, line_image, 1, 0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(lines_edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.7. Draw Hough lines from the edges in the masked region\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.7. Draw Hough lines from the edges in the masked region.png', dpi=300, bbox_inches='tight')
2.1.8. Superimpose the lines on the image to identify lane markings¶
Superimpose the extracted lines on the original image to potray the identification of lane markings in the image.
In [12]:
test_images = os.listdir("test_images/")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images/"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 1 # distance resolution in pixels of the Hough grid
theta = np.pi/180 # angular resolution in radians of the Hough grid
threshold = 40 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 10 #minimum number of pixels making up a line
max_line_gap = 70 # maximum gap in pixels between connectable line segments
line_image = np.copy(image)*0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(edges_masked, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
# Create a "color" binary image to combine with line image
color_edges = np.dstack((edges, edges, edges))
# Draw the lines on the edge image
lines_edges = cv2.addWeighted(color_edges, 0.8, line_image, 1, 0)
# Draw the lines on the copy of the origical image
image_lines = cv2.addWeighted(image, 0.8, line_image, 1, 0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image_lines, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('2.1.8. Superimpose the lines on the image to identify lane markings\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/2.1.8. Superimpose the lines on the image to identify lane markings.png', dpi=300, bbox_inches='tight')
-------- End of `2.1. Detect Lane Lines on images` --------
2.2 Detect Lane Lines in video stream¶
Import libraries to work on video streams¶
In [13]:
# Import everything needed to edit/save/watch video clips
from moviepy.editor import VideoFileClip
from IPython.display import HTML
You know what's cooler than drawing lanes over images? Drawing lanes over video!
We can test our solution on two provided videos and one optional challenge video:
solidWhiteRight.mp4, solidYellowLeft.mp4, and challenge.mp4
Note: if you get an import error when you run the next cell, try changing your kernel (select the Kernel menu above --> Change Kernel). Still have problems? Try relaunching Jupyter Notebook from the terminal prompt. Also, consult the forums for more troubleshooting tips.
If you get an error that looks like this:
NeedDownloadError: Need ffmpeg exe.
You can download it by calling:
imageio.plugins.ffmpeg.download()Follow the instructions in the error message and check out this forum post for more troubleshooting tips across operating systems.
In [14]:
HTML("""
<video width="480" height="270" controls>
<source src="{0}">
</video>
""".format("examples/raw-lines-example.mp4"))
Out[14]:
Pipeline¶
In [15]:
def process_image(read_image):
'''NOTE: The output you return should be a color image (3 channel) for processing video below'''
# read and make a copy of the image
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image = hough_lines(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1.0, β=1.0, γ=0.)
result = image_lanes
# return the final image where lines are drawn on lanes
return result
2.2.1. Detect Lane Lines in solidWhiteRight video¶
In [16]:
white_output = 'test_videos_output/1_1_solidWhiteRight_houghlines.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4").subclip(0,5)
clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4")
white_clip = clip1.fl_image(process_image) #NOTE: this function expects color images!!
%time white_clip.write_videofile(white_output, audio=False)
Play the video inline, or if you prefer find the video in your filesystem (should be in the same directory) and play it in your video player of choice.
In [17]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[17]:
2.2.2. Detect Lane Lines in solidYellowLeft video¶
In [18]:
yellow_output = 'test_videos_output/2_1_solidYellowLeft_houghlines.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4').subclip(0,5)
clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4')
yellow_clip = clip2.fl_image(process_image)
%time yellow_clip.write_videofile(yellow_output, audio=False)
In [19]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(yellow_output))
Out[19]:
2.2.3. Detect Lane Lines in challenge video¶
In [20]:
challenge_output = 'test_videos_output/3_1_challenge_houghlines.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
# clip3 = VideoFileClip('test_videos/challenge.mp4').subclip(0,4)
clip3 = VideoFileClip('test_videos/challenge.mp4')
challenge_clip = clip3.fl_image(process_image)
%time challenge_clip.write_videofile(challenge_output, audio=False)
In [21]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(challenge_output))
Out[21]:
The above videos depicts that the pipeline is able to identify the lane markings in the
1_1_solidWhiteRight_houghlinesand2_1_solidYellowLeft_houghlines, however it does not perform well on the3_1_challenge_houghlinesvideo as it contains a lot of nosie in the form of shadows, curves, and lane line fadeouts.
-------- End of `2.2. Detect Lane Lines in video stream` --------
3. Extrapolated Lane Line Detection¶
===================================================
Goal 2: Extrapolated Lane Detection Pipeline
At this point, we are successful with making the pipeline and tuning the parameters, and have the Hough line segments drawn onto the road, but what about identifying the full extent of the lane and marking it clearly as in the example video (P1_example.mp4)? Think about defining a line to run the full length of the visible lane based on the line segments you identified with the Hough Transform. The goal is to improvize the pipeline to detect the line segments in the image, then average/extrapolate them and draw them onto the image for display (as below).
|
The orginal image used to detect land lines on the road |
→ |
The output detecting line segments using helper functions |
Helper functions - modified¶
In [22]:
# Import everything needed to edit/save/watch video clips
from moviepy.editor import VideoFileClip
from IPython.display import HTML
In [23]:
from sklearn.linear_model import LinearRegression
reg = LinearRegression()
In [24]:
# Initiate global variables
r_x1 = 0
r_x2 = 0
r_x3 = 0
l_x1 = 0
l_x2 = 0
l_x3 = 0
def add_image_line(img, x1, y1, x2, y2, x3, color, thickness):
imshape = img.shape
x1 = int(np.round(x1, 0))
x2 = int(np.round(x2, 0))
x3 = int(np.round(x3, 0))
dx_r = int(np.round(x1 + (imshape[1]/100), 0))
dx_l = int(np.round(x1 - (imshape[1]/100), 0))
dy = int(np.round(y2 + (imshape[0]/50), 0))
y1 = int(np.round(y1, 0))
y2 = int(np.round(y2, 0))
# draw the lines
cv2.line(img, (x1, y1), (x2, y2), color, thickness)
cv2.line(img, (dx_r, y1), (x3, dy), color, thickness)
cv2.line(img, (dx_l, y1), (x3, dy), color, thickness)
def draw_lines_extrapolated(img, lines, color=[9, 219, 44], thickness=5):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
global r_x1
global r_x2
global r_x3
global l_x1
global l_x2
global l_x3
rc = np.array([])
lc = np.array([])
rx = np.array([])
ry = np.array([])
lx = np.array([])
ly = np.array([])
imshape = img.shape
for line in lines:
for x1,y1,x2,y2 in line:
center = [(x1+x2)/2, (y1+y2)/2]
try:
slope = (y2-y1)/(x2-x1)
except ZeroDivisionError:
slope = np.inf
if slope > 0.5 and slope < 10 and x1 > (.50*imshape[1]) and x2 > (.50*imshape[1]):
rc = np.append(rc, center)
rx = np.append(rx, [x1,x2])
ry = np.append(ry, [y1,y2])
elif slope < -0.5 and slope > -10 and x1 < (.50*imshape[1]) and x2 < (.50*imshape[1]):
lc = np.append(lc, center)
lx = np.append(lx, [x1,x2])
ly = np.append(ly, [y1,y2])
r_center = np.mean(rc, axis = 0)
l_center = np.mean(lc, axis = 0)
y1 = imshape[0]
y2 = imshape[0]*0.62
if not np.isnan(r_center).all():
rx = rx.reshape(-1,1)
reg.fit(rx, ry)
r_slope, r_intercept = reg.coef_[0], reg.intercept_
if r_slope > 0.5 and r_slope < 10:
r_x1 = (y1 - r_intercept) / r_slope
r_x2 = (y2 - r_intercept) / r_slope
r_x3 = ((y2 + (imshape[0]/50))- r_intercept) / r_slope
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
else:
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
elif np.isnan(r_center).all():
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
if not np.isnan(l_center).all():
lx = lx.reshape(-1,1)
reg.fit(lx, ly)
l_slope, l_intercept = reg.coef_[0], reg.intercept_
if l_slope < -0.5 and l_slope > -10:
l_x1 = (y1 - l_intercept) / l_slope
l_x2 = (y2 - l_intercept) / l_slope
l_x3 = ((y2 + (imshape[0]/50))- l_intercept) / l_slope
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
else:
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
elif np.isnan(l_center).all():
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
def hough_lines_extrapolated(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]),
minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
draw_lines_extrapolated(line_img, lines)
return line_img
3.1 Extrapolated Lane Line Detection on images¶
3.1.1. Display the images in the test images challenge directory¶
In [25]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(read_image)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.1. Display the images in the test images challenge directory\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.1. Display the images in the test images challenge directory.png', dpi=300, bbox_inches='tight')
3.1.2. Convert the images to grayscale (single channel)¶
As all the images in the test directory are colored (RGB channel), convert the images to single channel/grayscale for further computer vision operations.
In [26]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(gray, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.2. Convert the images to grayscale\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.2. Convert the images to grayscale.png', dpi=300, bbox_inches='tight')
3.1.3 Blur the grayscale image¶
In [27]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(blur_gray, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.3 Blur the grayscale image\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.3 Blur the grayscale image.png', dpi=300, bbox_inches='tight')
3.1.4. Extract the edges in the image¶
Extract the edges in the masked region by applying Canny Edge detection Computer Vision technique.
In [28]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.4. Extract the edges in the image\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.4. Extract the edges in the image.png', dpi=300, bbox_inches='tight')
3.1.5. Identify the region of interest¶
Identify the region of interest that is used for further operations. Take consideration of the hood of the car and eliminate it from the region of interest.
In [29]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# identify the region of interest
imshape = image.shape
x = [.55*imshape[1], .45*imshape[1], .15*imshape[1], .30*imshape[1],
.50*imshape[1], .70*imshape[1], .85*imshape[1], .55*imshape[1]]
y = [.60*imshape[0], .60*imshape[0], .90*imshape[0], .90*imshape[0],
.60*imshape[0], .90*imshape[0], .90*imshape[0], .60*imshape[0]]
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.plot(x, y, 'b--', lw=2)
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.5. Identify the region of interest\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.5. Identify the region of interest.png', dpi=300, bbox_inches='tight')
3.1.6. Mask the excess region in the images¶
Keep the region of interest that is used for further operations and remove/darken the rest of the image to ease the identification of the lanes ahead of the vehicle.
In [30]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# identify the region of interest
imshape = image.shape
x = [.55*imshape[1], .45*imshape[1], .15*imshape[1], .30*imshape[1],
.50*imshape[1], .70*imshape[1], .85*imshape[1], .55*imshape[1]]
y = [.60*imshape[0], .60*imshape[0], .90*imshape[0], .90*imshape[0],
.60*imshape[0], .90*imshape[0], .90*imshape[0], .60*imshape[0]]
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a custom polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(edges_masked, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.plot(x, y, 'b--', lw=2)
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.6. Mask the excess region in the images\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.6. Mask the excess region in the images.png', dpi=300, bbox_inches='tight')
3.1.7. Draw Hough lines from the edges in the masked region¶
Use Hough Line transformation to identify lines from the edges and draw the lines using various input parameters.
In [31]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# identify the region of interest
imshape = image.shape
x = [.55*imshape[1], .45*imshape[1], .15*imshape[1], .30*imshape[1],
.50*imshape[1], .70*imshape[1], .85*imshape[1], .55*imshape[1]]
y = [.60*imshape[0], .60*imshape[0], .90*imshape[0], .90*imshape[0],
.60*imshape[0], .90*imshape[0], .90*imshape[0], .60*imshape[0]]
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a custom polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 1 # distance resolution in pixels of the Hough grid
theta = np.pi/180 # angular resolution in radians of the Hough grid
threshold = 40 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 10 #minimum number of pixels making up a line
max_line_gap = 70 # maximum gap in pixels between connectable line segments
line_image = np.copy(image)*0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(edges_masked, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
# Create a "color" binary image to combine with line image
color_edges = np.dstack((edges_masked, edges_masked, edges_masked))
# Draw the lines on the edge image
lines_edges = cv2.addWeighted(color_edges, 0.8, line_image, 1, 0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(lines_edges, cmap='gray')
ax.set_title(test_image.split('.')[0])
ax.plot(x, y, 'b--', lw=2)
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.7. Draw Hough lines from the edges in the masked region\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.7. Draw Hough lines from the edges in the masked region.png', dpi=300, bbox_inches='tight')
3.1.8. Superimpose the lines on the image to identify lane markings¶
Superimpose the extracted lines on the original image to potray the identification of lane markings in the image.
In [32]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a custom polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 1 # distance resolution in pixels of the Hough grid
theta = np.pi/180 # angular resolution in radians of the Hough grid
threshold = 40 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 10 #minimum number of pixels making up a line
max_line_gap = 70 # maximum gap in pixels between connectable line segments
line_image = np.copy(image)*0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(edges_masked, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1,y1,x2,y2 in line:
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
# Draw the lines on the copy of the origical image
image_lines = cv2.addWeighted(image, 0.8, line_image, 1, 0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image_lines)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.8. Superimpose the lines on the image to identify lane markings\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.8. Superimpose the lines on the image to identify lane markings.png', dpi=300, bbox_inches='tight')
3.1.9. Filter the redundant lines in the images¶
Remove any lines with extreme slopes based on the position, that is left side or right side of the road in the image.
In [33]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
# Define a kernel size and apply Gaussian smoothing
kernel_size = 5
blur_gray = cv2.GaussianBlur(gray,(kernel_size, kernel_size),0)
# Define our parameters for Canny and apply
low_threshold = 50
high_threshold = 150
edges = cv2.Canny(blur_gray, low_threshold, high_threshold)
# Next we'll create a masked edges image using cv2.fillPoly()
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a custom polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
cv2.fillPoly(mask, vertices, ignore_mask_color)
edges_masked = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 1 # distance resolution in pixels of the Hough grid
theta = np.pi/180 # angular resolution in radians of the Hough grid
threshold = 40 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 10 #minimum number of pixels making up a line
max_line_gap = 70 # maximum gap in pixels between connectable line segments
line_image = np.copy(image)*0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(edges_masked, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1,y1,x2,y2 in line:
try:
slope = (y2-y1)/(x2-x1)
except ZeroDivisionError:
slope = np.inf
if slope > 0.5 and slope < 10 and x1 > (.50*imshape[1]) and x2 > (.50*imshape[1]):
print('Right Lane with slope: ',slope)
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
elif slope < -0.5 and slope > -10 and x1 < (.50*imshape[1]) and x2 < (.50*imshape[1]):
print('Left Lane with slope : ',slope)
cv2.line(line_image,(x1,y1),(x2,y2),(255,0,0),10)
# Draw the lines on the copy of the origical image
image_lines = cv2.addWeighted(image, 0.8, line_image, 1, 0)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image_lines)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.9. Filter the redundant lines in the images\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.9. Filter the redundant lines in the images.png', dpi=300, bbox_inches='tight')
3.1.10. Extrapoate lines with linear regression¶
Sort the lines based on the position (left lane, right lane) and perform a linear regression on the points based on their co-ordinates to fit the best line. Also perform the weighted addition of the slopes with previous slope with low learning rate to increase the stability of the line.
In [34]:
test_images = os.listdir("test_images_challenge")
# initiate figure and subplots
fig = plt.figure(figsize=[12,8])
nrows, ncols = 2, 2
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images_challenge"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image= hough_lines_extrapolated(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1, β=0.6, γ=0.)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image_lanes)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.10. Extrapoate lines with linear regression\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.10. Extrapoate lines with linear regression.png', dpi=300, bbox_inches='tight')
3.1.11. Extrapoate lines on test images set¶
In [35]:
test_images = os.listdir("test_images")
# initiate figure and subplots
fig = plt.figure(figsize=[20,8])
nrows, ncols = 2, 3
for i, test_image in enumerate(test_images, 1):
# construct the path to locate the image
root_directory = "test_images"
image_path = test_image
path = os.path.join(root_directory, image_path)
# read and make a copy of the image
read_image = mpimg.imread(path)
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image= hough_lines_extrapolated(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1, β=0.6, γ=0.)
# display the image
ax = fig.add_subplot(nrows, ncols, i)
ax.imshow(image_lanes)
ax.set_title(test_image.split('.')[0])
ax.axis('off');
# adjust subplots and add title
plt.subplots_adjust(wspace=0.2, hspace=0.3)
plt.subplots_adjust(top=0.85)
plt.suptitle('3.1.11. Extrapoate lines on test images set\n', fontsize = 16, weight = 'bold');
# savefig by passing (bbox_inches='tight'),which will adjust the figure to include all of the x and y labels
plt.savefig('test_images_output/3.1.11. Extrapoate lines on test images set.png', dpi=300, bbox_inches='tight')
Optional content walkthrough:
3.1.12 Linear Regression¶
How extraplation of line from points achieved with linear regression?In [36]:
def draw_lines_linear_regression(img, lines, color=[9, 219, 44], thickness=8):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
rx = []
ry = []
lx = []
ly = []
for line in lines:
for x1,y1,x2,y2 in line:
slope = (y2-y1)/(x2-x1)
center = [(x1+x2)/2, (y1+y2)/2]
if slope > 0.5 and slope < 10:
rx.extend([x1,x2])
ry.extend([y1,y2])
elif slope < -0.5 and slope > -10:
lx.extend([x1,x2])
ly.extend([y1,y2])
return rx, ry, lx, ly
def hough_lines_linear_regression(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]),
minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
rx, ry, lx, ly = draw_lines_linear_regression(line_img, lines)
return line_img, rx, ry, lx, ly
In [37]:
def process_image_linear_regression(read_image):
'''NOTE: The output you return should be a color image (3 channel) for processing video below'''
# read and make a copy of the image
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image, rx, ry, lx, ly = hough_lines_linear_regression(masked_edges, rho = 2, theta = np.pi/180,
threshold = 10, min_line_len = 40, max_line_gap = 20)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=0.8, β=1.0, γ=0.)
result = image_lanes
# return the final image where lines are drawn on lanes
return result, rx, ry, lx, ly
Obtain individual points related to left lane and right lanes for a test image:
In [38]:
root_directory = "test_images_challenge"
image_path = 'challengeImage1.jpg'
path = os.path.join(root_directory, image_path)
read_image= mpimg.imread(path)
plt.imshow(read_image);
In [39]:
image, rx, ry, lx, ly = process_image_linear_regression(read_image)
In [40]:
imshape = image.shape
imshape
Out[40]:
Right Lane:
In [41]:
rx, ry
Out[41]:
In [42]:
plt.imshow(image, alpha=0.5)
plt.scatter(rx, ry, color = 'red', s = 30, edgecolors='black');
In [43]:
plt.scatter(rx, ry, color = 'red');
Fit the Linear Regression:
In [44]:
rx = np.array(rx)
rx = rx.reshape(-1,1)
rx.shape
Out[44]:
In [45]:
reg.fit(rx, ry)
Out[45]:
In [46]:
plt.scatter(rx,ry, color = 'red')
plt.plot(rx, reg.predict(rx), color='blue');
In [47]:
reg.coef_, reg.intercept_
Out[47]:
In [48]:
y1 = imshape[0]
x1 = (y1 - reg.intercept_) / reg.coef_[0]
In [49]:
y2 = 0.65 * imshape[0]
x2 = (y2 - reg.intercept_) / reg.coef_[0]
In [50]:
plt.imshow(image, alpha = 0.6)
r_x = [x1, x2]
r_y = [y1, y2]
plt.plot(r_x, r_y, color = 'red');
Left Lane:
In [51]:
lx, ly
Out[51]:
In [52]:
plt.imshow(image, alpha=0.5)
plt.scatter(lx, ly, color = 'red', s = 30, edgecolors='black');
In [53]:
plt.scatter(lx, ly, color = 'red');
Fit the Linear Regression:
In [54]:
lx = np.array(lx)
lx = lx.reshape(-1,1)
lx.shape
Out[54]:
In [55]:
reg.fit(lx,ly)
Out[55]:
In [56]:
plt.scatter(lx,ly, color = 'red')
plt.plot(lx, reg.predict(lx), color='blue');
In [57]:
reg.coef_, reg.intercept_
Out[57]:
In [58]:
y1 = imshape[0]
x1 = (y1 - reg.intercept_) / reg.coef_[0]
In [59]:
y2 = 0.65 * imshape[0]
x2 = (y2 - reg.intercept_) / reg.coef_[0]
In [60]:
plt.imshow(image, alpha = 0.6)
l_x = [x1, x2]
l_y = [y1, y2]
plt.plot(l_x, l_y, color = 'red');
Right Lane and Left Lane combined:
In [61]:
plt.imshow(image)
plt.plot(l_x, l_y, color = 'red');
plt.plot(r_x, r_y, color = 'red');
-------- End of `3.1. Extrapolated Lane Line Detection on images` --------
3.2 Extrapolated Lane Line Detection on videos¶
Goal: An example of the desired output¶
In [62]:
HTML("""
<video width="480" height="270" controls>
<source src="{0}">
</video>
""".format("examples/P1_example.mp4"))
Out[62]:
Pipeline¶
In [63]:
def process_image_extrapolated(read_image):
'''NOTE: The output you return should be a color image (3 channel) for processing video below'''
# read and make a copy of the image
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image= hough_lines_extrapolated(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1, β=0.6, γ=0.)
result = image_lanes
# return the final image where lines are drawn on lanes
return result
3.2.1. Detect extrapolated Lane Lines in solidWhiteRight video¶
In [64]:
white_output = 'test_videos_output/1_2_solidWhiteRight_extrapolated.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4").subclip(0,5)
clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4")
white_clip = clip1.fl_image(process_image_extrapolated) #NOTE: this function expects color images!!
%time white_clip.write_videofile(white_output, audio=False)
Play the video inline, or if you prefer find the video in your filesystem (should be in the same directory) and play it in your video player of choice.
In [65]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[65]:
3.2.2. Detect extrapolated Lane Lines in solidYellowLeft video¶
In [66]:
yellow_output = 'test_videos_output/2_2_solidYellowLeft_extrapolated.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4').subclip(0,5)
clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4')
yellow_clip = clip2.fl_image(process_image_extrapolated)
%time yellow_clip.write_videofile(yellow_output, audio=False)
In [67]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(yellow_output))
Out[67]:
3.2.3. Detect extrapolated Lane Lines in challenge video¶
In [68]:
challenge_output = 'test_videos_output/3_2_challenge_extrapolated.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
# clip3 = VideoFileClip('test_videos/challenge.mp4').subclip(0,4)
clip3 = VideoFileClip('test_videos/challenge.mp4')
challenge_clip = clip3.fl_image(process_image_extrapolated)
%time challenge_clip.write_videofile(challenge_output, audio=False)
In [69]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(challenge_output))
Out[69]:
The above videos depicts that the pipeline is able to identify the lane markings and draw the extrapolated lines in the videos
1_2_solidWhiteRight_houghlinesand2_2_solidYellowLeft_houghlines. It also performs well on the3_2_challenge_houghlinesvideo because of the modifications in the helper functions.
-------- End of `3.2. Extrapolated Lane Line Detection on videos` --------
4. Stabilize the Extrapolated Lane Line Detection¶
========================================================================
Stabilize the extrapolated lines by defining a learning rate with the previously detected lines.
Helper Functions Extrapolated and Stabilized:
In [70]:
# Initiate global variables
r_x1 = 0
r_x2 = 0
r_x3 = 0
l_x1 = 0
l_x2 = 0
l_x3 = 0
def add_image_lines(img, x1, y1, x2, y2, x3, color, thickness):
imshape = img.shape
x1 = int(np.round(x1, 0))
x2 = int(np.round(x2, 0))
x3 = int(np.round(x3, 0))
dx_r = int(np.round(x1 + (imshape[1]/100), 0))
dx_l = int(np.round(x1 - (imshape[1]/100), 0))
dy = int(np.round(y2 + (imshape[0]/50), 0))
y1 = int(np.round(y1, 0))
y2 = int(np.round(y2, 0))
# draw the lines
cv2.line(img, (x1, y1), (x2, y2), color, thickness)
cv2.line(img, (dx_r, y1), (x3, dy), color, thickness)
cv2.line(img, (dx_l, y1), (x3, dy), color, thickness)
# fill the lines
ignore_mask_color = color
vertices = np.array([[(dx_r, y1),
(dx_l, y1),
(x2, y2)]], dtype = np.int32)
cv2.fillPoly(img, vertices, ignore_mask_color)
def draw_lines_extrapolated_stable(img, lines, color=[9, 219, 44], thickness=5):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
global r_x1
global r_x2
global r_x3
global l_x1
global l_x2
global l_x3
rc = np.array([])
lc = np.array([])
rx = np.array([])
ry = np.array([])
lx = np.array([])
ly = np.array([])
imshape = img.shape
for line in lines:
for x1,y1,x2,y2 in line:
center = [(x1+x2)/2, (y1+y2)/2]
try:
slope = (y2-y1)/(x2-x1)
except ZeroDivisionError:
slope = np.inf
if slope > 0.5 and slope < 10 and x1 > (.50*imshape[1]) and x2 > (.50*imshape[1]):
rc = np.append(rc, center)
rx = np.append(rx, [x1,x2])
ry = np.append(ry, [y1,y2])
elif slope < -0.5 and slope > -10 and x1 < (.50*imshape[1]) and x2 < (.50*imshape[1]):
lc = np.append(lc, center)
lx = np.append(lx, [x1,x2])
ly = np.append(ly, [y1,y2])
r_center = np.mean(rc, axis = 0)
l_center = np.mean(lc, axis = 0)
y1 = imshape[0]
y2 = imshape[0]*0.65
if not np.isnan(r_center).all():
rx = rx.reshape(-1,1)
reg.fit(rx, ry)
r_slope, r_intercept = reg.coef_[0], reg.intercept_
if r_slope > 0.5 and r_slope < 10:
r_x1_new = (y1 - r_intercept) / r_slope
r_x2_new = (y2 - r_intercept) / r_slope
r_x3_new = ((y2 + (imshape[0]/50)) - r_intercept) / r_slope
if r_x1 == 0:
learning_rate = 1
else:
learning_rate = 0.2
r_x1 = (learning_rate * r_x1_new) + ((1 - learning_rate) * r_x1)
r_x2 = (learning_rate * r_x2_new) + ((1 - learning_rate) * r_x2)
r_x3 = (learning_rate * r_x3_new) + ((1 - learning_rate) * r_x3)
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
else:
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
elif np.isnan(r_center).all():
add_image_line(img, r_x1, y1, r_x2, y2, r_x3, color, thickness)
if not np.isnan(l_center).all():
lx = lx.reshape(-1,1)
reg.fit(lx, ly)
l_slope, l_intercept = reg.coef_[0], reg.intercept_
if l_slope < -0.5 and l_slope > -10:
l_x1_new = (y1 - l_intercept) / l_slope
l_x2_new = (y2 - l_intercept) / l_slope
l_x3_new = ((y2 + (imshape[0]/50)) - l_intercept) / l_slope
if l_x1 == 0:
learning_rate = 1
else:
learning_rate = 0.2
l_x1 = (learning_rate * l_x1_new) + ((1 - learning_rate) * l_x1)
l_x2 = (learning_rate * l_x2_new) + ((1 - learning_rate) * l_x2)
l_x3 = (learning_rate * l_x3_new) + ((1 - learning_rate) * l_x3)
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
else:
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
elif np.isnan(l_center).all():
add_image_line(img, l_x1, y1, l_x2, y2, l_x3, color, thickness)
def hough_lines_extrapolated_stable(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]),
minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
draw_lines_extrapolated_stable(line_img, lines)
return line_img
In [71]:
def process_image_extrapolated_stable(read_image):
'''NOTE: The output you return should be a color image (3 channel) for processing video below'''
# read and make a copy of the image
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image= hough_lines_extrapolated_stable(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1, β=0.6, γ=0.)
result = image_lanes
# return the final image where lines are drawn on lanes
return result
4.1.1. Stabilize the extrapolated Lane Lines in solidWhiteRight video¶
In [72]:
white_output = 'test_videos_output/1_3_solidWhiteRight_extrapolated_stable.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4").subclip(0,5)
clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4")
white_clip = clip1.fl_image(process_image_extrapolated_stable) #NOTE: this function expects color images!!
%time white_clip.write_videofile(white_output, audio=False)
Play the video inline, or if you prefer find the video in your filesystem (should be in the same directory) and play it in your video player of choice.
In [73]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[73]:
4.1.2. Stabilize extrapolated Lane Lines in solidYellowLeft video¶
In [74]:
yellow_output = 'test_videos_output/2_3_solidYellowLeft_extrapolated_stable.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4').subclip(0,5)
clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4')
yellow_clip = clip2.fl_image(process_image_extrapolated_stable)
%time yellow_clip.write_videofile(yellow_output, audio=False)
In [75]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(yellow_output))
Out[75]:
4.1.3. Stabilize the extrapolated Lane Lines in challenge video¶
In [76]:
challenge_output = 'test_videos_output/3_3_challenge_extrapolated_stable.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
# clip3 = VideoFileClip('test_videos/challenge.mp4').subclip(0,4)
clip3 = VideoFileClip('test_videos/challenge.mp4')
challenge_clip = clip3.fl_image(process_image_extrapolated_stable)
%time challenge_clip.write_videofile(challenge_output, audio=False)
In [77]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(challenge_output))
Out[77]:
-------- End of `4. Stabilize the Extrapolated Lane Line Detection` --------
5. Fill the Lane Line Detection¶
========================================================================
After Stabilizing the extrapolated lines by defining a learning rate with the previously detected lines, fill the coordinates between the left and right lane using poly fill.
Helper Functions for poly fill:
In [78]:
r_x1 = 0
r_x2 = 0
r_x3 = 0
l_x1 = 0
l_x2 = 0
l_x3 = 0
def fill_image_line(img, r_x1, r_x2, l_x1, l_x2, y1, y2, ignore_mask_color):
vertices = np.array([[(r_x1, y1),
(l_x1, y1),
(l_x2, y2),
(r_x2, y2)]], dtype = np.int32)
cv2.fillPoly(img, vertices, ignore_mask_color)
def draw_lines_improv2(img, lines, color=[9, 219, 44], thickness=5):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
global r_x1
global r_x2
global r_x3
global l_x1
global l_x2
global l_x3
rc = np.array([])
lc = np.array([])
rx = np.array([])
ry = np.array([])
lx = np.array([])
ly = np.array([])
imshape = img.shape
for line in lines:
for x1,y1,x2,y2 in line:
center = [(x1+x2)/2, (y1+y2)/2]
try:
slope = (y2-y1)/(x2-x1)
except ZeroDivisionError:
slope = np.inf
if slope > 0.5 and slope < 10 and x1 > (.50*imshape[1]) and x2 > (.50*imshape[1]):
rc = np.append(rc, center)
rx = np.append(rx, [x1,x2])
ry = np.append(ry, [y1,y2])
elif slope < -0.5 and slope > -10 and x1 < (.50*imshape[1]) and x2 < (.50*imshape[1]):
lc = np.append(lc, center)
lx = np.append(lx, [x1,x2])
ly = np.append(ly, [y1,y2])
r_center = np.mean(rc, axis = 0)
l_center = np.mean(lc, axis = 0)
y1 = imshape[0]
y2 = imshape[0]*0.65
if not np.isnan(r_center).all():
rx = rx.reshape(-1,1)
reg.fit(rx, ry)
r_slope, r_intercept = reg.coef_[0], reg.intercept_
if r_slope > 0.5 and r_slope < 10:
r_x1 = (y1 - r_intercept) / r_slope
r_x2 = (y2 - r_intercept) / r_slope
if not np.isnan(l_center).all():
lx = lx.reshape(-1,1)
reg.fit(lx, ly)
l_slope, l_intercept = reg.coef_[0], reg.intercept_
if l_slope < -0.5 and l_slope > -10:
l_x1 = (y1 - l_intercept) / l_slope
l_x2 = (y2 - l_intercept) / l_slope
fill_image_line(img, r_x1, r_x2, l_x1, l_x2, y1, y2, color)
def hough_lines_improv2(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]),
minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
draw_lines_improv2(line_img, lines)
return line_img
In [79]:
# Initiate global variables
r_x1 = 0
r_x2 = 0
r_x3 = 0
l_x1 = 0
l_x2 = 0
l_x3 = 0
def add_image_fill(img, r_x1, r_x2, l_x1, l_x2, y1, y2, ignore_mask_color):
vertices = np.array([[(r_x1, y1),
(l_x1, y1),
(l_x2, y2),
(r_x2, y2)]], dtype = np.int32)
cv2.fillPoly(img, vertices, ignore_mask_color)
def draw_lines_fill(img, lines, color=[9, 219, 44], thickness=5):
"""
NOTE: this is the function you might want to use as a starting point once you want to
average/extrapolate the line segments you detect to map out the full
extent of the lane (going from the result shown in raw-lines-example.mp4
to that shown in P1_example.mp4).
Think about things like separating line segments by their
slope ((y2-y1)/(x2-x1)) to decide which segments are part of the left
line vs. the right line. Then, you can average the position of each of
the lines and extrapolate to the top and bottom of the lane.
This function draws `lines` with `color` and `thickness`.
Lines are drawn on the image inplace (mutates the image).
If you want to make the lines semi-transparent, think about combining
this function with the weighted_img() function below
"""
global r_x1
global r_x2
global r_x3
global l_x1
global l_x2
global l_x3
rc = np.array([])
lc = np.array([])
rx = np.array([])
ry = np.array([])
lx = np.array([])
ly = np.array([])
imshape = img.shape
for line in lines:
for x1,y1,x2,y2 in line:
center = [(x1+x2)/2, (y1+y2)/2]
try:
slope = (y2-y1)/(x2-x1)
except ZeroDivisionError:
slope = np.inf
if slope > 0.5 and slope < 10 and x1 > (.50*imshape[1]) and x2 > (.50*imshape[1]):
rc = np.append(rc, center)
rx = np.append(rx, [x1,x2])
ry = np.append(ry, [y1,y2])
elif slope < -0.5 and slope > -10 and x1 < (.50*imshape[1]) and x2 < (.50*imshape[1]):
lc = np.append(lc, center)
lx = np.append(lx, [x1,x2])
ly = np.append(ly, [y1,y2])
r_center = np.mean(rc, axis = 0)
l_center = np.mean(lc, axis = 0)
y1 = imshape[0]
y2 = imshape[0]*0.65
if not np.isnan(r_center).all():
rx = rx.reshape(-1,1)
reg.fit(rx, ry)
r_slope, r_intercept = reg.coef_[0], reg.intercept_
if r_slope > 0.5 and r_slope < 10:
r_x1_new = (y1 - r_intercept) / r_slope
r_x2_new = (y2 - r_intercept) / r_slope
r_x3_new = ((y2 + (imshape[0]/50)) - r_intercept) / r_slope
if r_x1 == 0:
learning_rate = 1
else:
learning_rate = 0.2
r_x1 = (learning_rate * r_x1_new) + ((1 - learning_rate) * r_x1)
r_x2 = (learning_rate * r_x2_new) + ((1 - learning_rate) * r_x2)
r_x3 = (learning_rate * r_x3_new) + ((1 - learning_rate) * r_x3)
if not np.isnan(l_center).all():
lx = lx.reshape(-1,1)
reg.fit(lx, ly)
l_slope, l_intercept = reg.coef_[0], reg.intercept_
if l_slope < -0.5 and l_slope > -10:
l_x1_new = (y1 - l_intercept) / l_slope
l_x2_new = (y2 - l_intercept) / l_slope
l_x3_new = ((y2 + (imshape[0]/50)) - l_intercept) / l_slope
if l_x1 == 0:
learning_rate = 1
else:
learning_rate = 0.2
l_x1 = (learning_rate * l_x1_new) + ((1 - learning_rate) * l_x1)
l_x2 = (learning_rate * l_x2_new) + ((1 - learning_rate) * l_x2)
l_x3 = (learning_rate * l_x3_new) + ((1 - learning_rate) * l_x3)
# fill the image
add_image_fill(img, r_x1, r_x2, l_x1, l_x2, y1, y2, color)
def hough_lines_fill(img, rho, theta, threshold, min_line_len, max_line_gap):
"""
`img` should be the output of a Canny transform.
Returns an image with hough lines drawn.
"""
lines = cv2.HoughLinesP(img, rho, theta, threshold, np.array([]),
minLineLength=min_line_len, maxLineGap=max_line_gap)
line_img = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
draw_lines_fill(line_img, lines)
return line_img
In [80]:
def process_image_fill(read_image):
'''NOTE: The output you return should be a color image (3 channel) for processing video below'''
# read and make a copy of the image
image = np.copy(read_image)
# convert the image to grayscale
gray_image = grayscale(image)
# Define a kernel size and apply Gaussian smoothing
blur_gray = gaussian_blur(gray_image, kernel_size = 5)
# Define our parameters for Canny and apply
edges = canny(blur_gray, low_threshold = 50, high_threshold = 150)
# maske the region of interest
mask = np.zeros_like(edges)
ignore_mask_color = 255
# This time we are defining a four sided polygon to mask
imshape = image.shape
vertices = np.array([[(.55*imshape[1], .60*imshape[0]),
(.45*imshape[1], .60*imshape[0]),
(.15*imshape[1], .90*imshape[0]),
(.30*imshape[1], .90*imshape[0]),
(.50*imshape[1], .60*imshape[0]),
(.70*imshape[1], .90*imshape[0]),
(.85*imshape[1], .90*imshape[0]),
(.55*imshape[1], .60*imshape[0])]], dtype=np.int32)
masked_edges = region_of_interest(edges, vertices)
# draw lines using hough transform
line_image= hough_lines_fill(masked_edges, rho = 1, theta = np.pi/180,
threshold = 40, min_line_len = 10, max_line_gap = 70)
# perform weighted addition of line_image and original image to potray lane markings
image_lanes = weighted_img(line_image, image, α=1, β=0.6, γ=0.)
result = image_lanes
# return the final image where lines are drawn on lanes
return result
5.1.1. Fill the space between the Lane Lines in solidWhiteRight video¶
In [81]:
white_output = 'test_videos_output/1_4_solidWhiteRight_fill.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4").subclip(0,5)
clip1 = VideoFileClip("test_videos/solidWhiteRight.mp4")
white_clip = clip1.fl_image(process_image_fill) #NOTE: this function expects color images!!
%time white_clip.write_videofile(white_output, audio=False)
Play the video inline, or if you prefer find the video in your filesystem (should be in the same directory) and play it in your video player of choice.
In [82]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[82]:
5.1.2. Fill the space between the Lane Lines in solidYellowLeft video¶
In [83]:
yellow_output = 'test_videos_output/2_4_solidYellowLeft_fill.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
##clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4').subclip(0,5)
clip2 = VideoFileClip('test_videos/solidYellowLeft.mp4')
yellow_clip = clip2.fl_image(process_image_fill)
%time yellow_clip.write_videofile(yellow_output, audio=False)
In [84]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(yellow_output))
Out[84]:
5.1.3. Fill the space between the Lane Lines in challenge video¶
In [85]:
challenge_output = 'test_videos_output/3_4_challenge_fill.mp4'
## To speed up the testing process you may want to try your pipeline on a shorter subclip of the video
## To do so add .subclip(start_second,end_second) to the end of the line below
## Where start_second and end_second are integer values representing the start and end of the subclip
## You may also uncomment the following line for a subclip of the first 5 seconds
# clip3 = VideoFileClip('test_videos/challenge.mp4').subclip(0,4)
clip3 = VideoFileClip('test_videos/challenge.mp4')
challenge_clip = clip3.fl_image(process_image_fill)
%time challenge_clip.write_videofile(challenge_output, audio=False)
In [86]:
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(challenge_output))
Out[86]:
-------- End of `5. Fill the Lane Line Detection` --------
|
-------- End of Notebook --------
|
|